Video Abstracting at a Semantical Level

One the most common form of a video abstract is the movie trailer. Contemporary movie trailers share a common structure across genres which allows for an automatic generation and also reflects the corresponding moviea s composition. In this thesis a system for the automatic generation of trailers is presented. In addition to action trailers, the system is able to deal with further genres such as Horror and comedy trailers, which were first manually analyzed in order to identify their basic structures. To simplify the modeling of trailers and the abstract generation itself a new video abstracting application was developed. This application is capable of performing all steps of the abstract generation automatically and allows for previews and manual optimizations. Based on this system, new abstracting models for horror and comedy trailers were created and the corresponding trailers have been automatically generated using the new abstracting models. In an evaluation the automatic trailers were compared to the original Trailers and showed a similar structure. However, the automatically generated trailers still do not exhibit the full perfection of the Hollywood originals as they lack intentional storylines across shots.